Companies Are Rationing AI Tokens — SMBs Can Learn From The Waste

Remember six months ago, when every tech CEO was bragging about how much their teams were using AI? Leaderboards tracking how many tokens each employee burned. Internal competitions to see who could prompt ChatGPT the most. Articles celebrating the “tokenmaxxing” era.

Yeah. That was stupid. And the bill just came due.

TechCrunch reported on June 24 that companies are now scrambling to stop employees from maxing out AI budgets with trivial tasks. The same organizations that built usage leaderboards are now implementing AI budget caps, per-employee token limits, and strict usage policies. The “tokenmaxxing” era is dead. We’re entering the era of token rationing.

For SMBs, this is both a warning and an opportunity. You don’t have the margin to waste $5,000 a month on employees asking AI to rewrite the same email eight different ways. But you also don’t want to overcorrect and block the AI usage that actually drives results. The companies currently panicking over AI budgets are proof that the tools work — they just need better governance around how they’re deployed.

The difference between AI spend that pays off and AI spend that burns cash comes down to one thing: tracking ROI. And for most SMBs, that tracking just doesn’t exist yet. Most owners can tell you their total AI subscription cost but can’t tell you what percentage of that spend is producing revenue, improving efficiency, or doing nothing at all.

How We Got Here: The Tokenmaxxing Hangover

The AI industry spent 2025 and early 2026 encouraging companies to max out their AI budgets. Companies built employee leaderboards to encourage internal AI usage. Amazon reportedly had a developer leaderboard for AI usage that the company eventually deleted to minimize costs — but by then the habits were formed.

The result: employees using expensive AI tools for tasks that didn’t need them. Asking GPT-4 to summarize a two-sentence email. Running Claude to draft a grocery list. Generating five different versions of a simple Slack message because the first one was “close but not quite right.”

One company cited by TechCrunch discovered that over 40% of their AI spend went to tasks that employees had previously handled without AI — and handled just fine. The AI wasn’t making them faster. It was making them lazier. Instead of writing a five-sentence email themselves, they’d prompt an AI, review the output, tweak it, and spend the same amount of time they would have spent writing it directly — just with an extra $0.50 of compute consumed every time.

TechCrunch’s analysis frames this as the natural arc of enterprise AI adoption: companies try blanket access → usage explodes → costs explode → panic → rationing begins. Most large enterprises are currently between “panic” and “rationing.”

The New York Times ran a piece on June 18 tracing the minimization trend, noting that companies spent heavily on AI earlier in the year without measuring whether the output was worth the cost. Now they’re cutting back hard. One Fortune 500 company cited in the article implemented a $50-per-month-per-employee AI budget cap. That’s barely enough for heavy daily usage of Claude or GPT-4.

Why This Hits SMBs Harder

Large enterprises can absorb a $50,000 AI overspend and call it a learning expense. A small business with 20 employees and $3 million in revenue cannot.

For American SMBs, every dollar of AI spend that doesn’t produce measurable output is a dollar that could have gone to hiring, marketing, or inventory. The margin for error is nonexistent. And yet, many SMBs have the same blanket-access approach that got enterprises into trouble — everyone gets a ChatGPT subscription, turn it on, figure it out later.

The difference is that when an enterprise overspends, they hire a consultant to fix it. When an SMB overspends, they either cancel everything and miss the real benefits, or they keep burning money and wonder why margins are shrinking.

The third option — tracking ROI and optimizing spend — is the one most SMBs ignore. And it’s the one that actually works.

How to Track AI ROI Without a Data Team

You don’t need a dedicated AI budget analyst. You need a system that takes 30 minutes to set up and saves you 10-20% of your monthly AI spend.

Step 1: Audit your current AI subscriptions. Pull a list of every AI tool your team uses and what each costs per month. Include ChatGPT, Claude, Copilot, Grammarly, Canva AI, Notion AI — everything. Most SMBs are shocked by the total.

Step 2: Categorize usage. Not all AI usage is equal. Split it into three buckets: revenue-generating (drafting proposals, analyzing data, creating client-facing content), productivity-enhancing (summarizing meetings, drafting internal documents, automating admin work), and discretionary (recreational prompts, over-optimizing low-value output).

Step 3: Survey your team for 5 minutes. Ask each person: “What AI tools do you use? What do you use them for? What would you lose if you couldn’t use them tomorrow?” The answers will tell you exactly what’s essential and what’s waste.

Step 4: Set per-tool budgets. Assign a monthly budget to each AI tool based on actual business need. ChatGPT at $20/head is fine if it’s generating revenue. ChatGPT at $200/head because someone is generating 47 variations of an internal memo is not fine.

Step 5: Review monthly. Compare AI spend against measurable outcomes. Did the team close more deals? Produce more content? Reduce time on specific tasks? If you can’t connect spend to a result, cut it.

What Smart AI Budgeting Looks Like

The companies that survive the token rationing era aren’t the ones that ban AI entirely — they’re the ones that deploy it surgically with clear targeting and measurable outcomes.

Instead of giving every employee unlimited access to the most expensive model, smart SMBs are:

Using tiered access. The sales team gets GPT-4 or Claude Opus because they’re drafting client proposals. The admin team gets a smaller, cheaper model for internal tasks. Different tools for different jobs, priced accordingly.

Setting task boundaries. AI for research, drafting, and analysis = approved. AI for rewriting the same sentence 12 times = not approved. Clear policies eliminate the gray area where most waste lives.

Measuring output, not usage. The companies that failed tracked how many tokens their teams used. The companies that succeed track what those tokens produced. Token counts are a vanity metric. Output is the only metric that matters.

Using free and low-cost alternatives for low-value work. Not every task needs the model with the highest benchmark score. A 7B-parameter model running locally can summarize an email thread just as well as GPT-4 for a fraction of the cost. The smart move is matching the tool to the task.

The Real Cost of Untracked AI Spend

Let’s run the numbers for a 15-person SMB.

Fifteen ChatGPT Team subscriptions at $25/month each = $375/month. Six Claude Team subscriptions at $30/month each = $180/month. Three Copilot Pro subscriptions at $20/month each = $60/month. Canva AI, Grammarly Premium, and three other AI tools at $15-50/month each = roughly $150/month.

Total: roughly $765/month in baseline AI spend. That’s $9,180 per year.

Now add the inefficiency layer. If even 30% of that usage is low-value — rewriting emails, generating content nobody uses, asking questions that could be answered with a 5-second Google search — that’s $2,754 in annual waste. For a 15-person business, that’s real money.

Three months of untracked AI waste would cover a weekend marketing push that could generate ten times that in revenue. Five months would pay for a new software subscription that actually drives growth. The opportunity cost of unchecked AI spending is higher than most owners realize because the waste is invisible — it just looks like normal expenses on a credit card statement.

Here’s another angle worth considering. Gartner’s 2026 enterprise AI spending forecast showed that organizations spent 40% more on AI tools in Q1 than they budgeted. Of that overspend, roughly 35% went to low-value usage — tasks that could have been handled by cheaper alternatives or eliminated entirely. That’s not just an enterprise problem. SMBs running a $1,000 monthly AI spend are likely burning $350 of it on things that don’t move the needle.

The Information reported that Meta was one of the first companies to address this directly, implementing internal AI policies that prioritized high-value use cases and restricted discretionary token consumption. The approach was simple: if a task doesn’t connect to a measurable business outcome, it doesn’t get AI budget. That principle scales from a Fortune 500 boardroom to a 10-person shop without modification.

Extrapolate that across a 50-person company and you’re looking at $15,000-20,000 in annual AI waste — before accounting for the more expensive enterprise plans that larger companies typically use. That’s not a rounding error. For many SMBs, that’s the difference between a profitable year and a tight one.

The SMB Advantage

Here’s the irony: SMBs are actually in a better position to manage AI spend than large enterprises. You have fewer people to audit, simpler tool stacks, and more direct visibility into whether AI is actually producing results.

A 20-person company can audit its entire AI spend in an afternoon. A Fortune 500 company needs a dedicated team running for months to get a similar picture. That speed is the SMB advantage. Use it.

The companies that figure out AI ROI first will have a structural advantage over their competitors. While larger organizations are still debating whether to buy or build an internal AI tracking dashboard, you can have your entire AI budget optimized and running by the end of the week.

FAQ

What is token rationing?

Token rationing is the practice of limiting how much AI usage employees can generate, typically through per-user budget caps, monthly token limits, or restricted access to expensive models. It’s the emerging counter-trend to the “tokenmaxxing” era.

How can SMBs track AI ROI without a data team?

Audit all current AI subscriptions, categorize usage into revenue-generating vs discretionary, survey your team about which tools they actually need, set per-tool budgets, and review monthly against measurable outcomes.

How much AI waste is typical for a small business?

For a 15-person SMB with baseline AI subscriptions of ~$765/month, roughly 30% ($2,754/year) may be low-value usage. For a 50-person company, waste can reach $15,000-20,000 per year.

Should SMBs ban AI to control costs?

No. The solution isn’t banning AI — it’s deploying it surgically with tiered access, task boundaries, and output-based measurement. Companies that track ROI outperform those that either spend indiscriminately or ban AI entirely.